Machine Learning based Social Network Data Analysis and Prediction for Wireless Communication Network Optimization

Due to the rapid development of the wireless communication network, the total amount of data in the future is expected to triple. In the next decade, its total will grow by a factor of 1000, especially in the field of wireless communication networks. With the popularity of mobile devices and the rapid development of multimedia applications such as social networking, video sharing, and telepresence, mobile communication networks have become an integral part of people's daily lives. With in-depth research, researchers wish to find that through mining analysis, effective information and patterns can be found from mobile traffic and social network data to optimize wireless networks. This also echoes the main elaboration and research of this thesis.

Firstly, based on the geographical information of the collected Twitter traffic, the density-based noisy application spatial clustering (DBSCAN) is a traffic distribution that is relatively suitable for reality by comparing clustering algorithms. Then, a framework based on social network data that identifies and clusters mobile traffic hotspots using the DBSCAN algorithm is proposed. By comparing the hotspots of the cluster with the existing macro base station (MBSs) locations, it can be determined whether other small base stations (SBSs) need to be deployed and that such deployments can effectively improve the quality of service (QoS).

On the other hand, research focuses on the temporal and spatial dimensions of data. After partitioning the data into the grid, the region can be classified by the functions it contains. A Twitter traffic prediction framework is proposed, which aggregates geographic regions with similar traffic patterns into a group and predicts the Twitter traffic long-term and short-term memory (LSTM) network for each regional group based on each group of geographic regions. The proposed framework not only allows multiple regions to share the same LSTM prediction model but also extends the training set of each model, thereby reducing the risk of over-fitting during training.

With the study of social network data, some special crowding events will cause the population to aggregate in a small area. This phenomenon will lead to a significant increase in the total amount of traffic over a period. Thus, in another research direction. The areas affected by special crowding events by detecting and classifying Twitter traffic. Then train and test through the traffic pattern in these areas. The result is positive. The framework based on the dynamic time warping (DTW) algorithm can well determine whether an area is affected by a certain crowding event.